Purpose of selection operators

[Ramon Diaz-Uriarte]

Dear Ken,

> Qn:
> 1) Are my observation correct ? Does both cases always apply? ( plz
> elaborate when it does and doesn't )

WIth non-overlapping generations (e.g., SImpleGA in galib) the second is not an
issue, since the whole population is replaced. 

As far as I can tell (and, mind you, I am a zoologist, not a computer
scientist) in GA's with overlapping generations (such as the steady state GAs in
galib) an individual's reproduction coincides with its death; in other words,
an individual reproduces when it is replaced from the population. This is also
the way some biological systems are modeled using difference equations (such as
the discrete time, overallping generations replicator dynamics).  Thus, with
overalpping generations, individuals with higher fitness are more likely not to
be replaced (i.e., to be present in the following generation). However, this
can be at odds with what goes on in many biological systems, where individuals
with higher fitness can both survive for longer periods, and leave more
offspring while they are leaving.

> 
> 2) Are the main steps selection - crossover - mutation -> New pop.
> always in this order?

That is what you can see in the fig. in p. 6 of the galib documentation. 
As I understand, for practical purposes the following are almost equivalent:

selection-> crossover->mutation->New pop;

selection-> mutation -> crossover -> New. pop.

In both cases the end result is that the offspring might not be identical to
the parents because of a change due to mutation.  The slight difference between
the two is that in the second case,  the offspring would share the same
mutation; with the first diagram, offspring might differ w.r.t. mutations,
because each can mutate independently. (Yes, biologically, things happen
as in the second diagram.)


Hope this helps
> 
>     TIA!
> 
> ken
-- 
Ramón Díaz-Uriarte
Triana 47
28016 Madrid
Spain

email:ramon-diaz at teleline.es
Phone: +-34-918-513-966
       +-34-657-186-407